Author: Bhavesh Patel
Background: Retinal detachment is important to diagnose quickly in the Emergency Department (ED) due to its potential for causing vision loss. However, in busy EDs, the challenging and time-consuming nature of the traditional fundoscopic ocular exam can delay diagnoses. Research has shown readily available ocular point-of-care ultrasound (POCUS) may be a more focused and timely alternative for diagnosing retinal detachment. However, the accuracy of diagnosis from these scans depends highly on the skill of sonographer. Artificial intelligence (AI) has been used to improve scan accuracy. Google Deepmind is currently researching machine learning to interpret optical coherence tomography (OCT), an echo technique examining the fundus, in diagnosing diabetic retinopathy and age-related macular degeneration (AMD). We propose to use analogous machine learning to analyze the sensitivity and specificity of computer-aided diagnosis (CAD) of eye complaints in the ED by ocular POCUS as compared to the gold standard of ophthalmology diagnosis via fundoscopic exam. Ideas: The primary objective of the proposed study will be to utilize machine learning to interpret ocular POCUS for the diagnosis of retinal detachment, vitreous hemorrhage, and vitreous detachment in patients seen in the ED. Machine learning in ultrasound machines can help ED sonographers, who may not be trained specifically for ocular POCUS, to confirm that they are capturing the correct image of the fundus. CAD can be programmed into ultrasound machines to serve as an additional screening technique for diagnosing eye complaints. This will be a prospective, cross-sectional study that will compare the computer diagnoses with that of the Ophthalmologist, who will be blinded to the machine’s diagnosis from POCUS. Sonographers will be Emergency Medicine residents and attending physicians. It is important to recognize that POCUS has proven in and of itself to be a viable means of diagnosing retinal detachments by ophthalmologists since 1969. The intent of this study is to further develop ultrasound by means of CAD to better the quality and rate of assessment for patients with ocular complaints by giving a diagnosis that will accurately represent one given by an Ophthalmologist. If this study proves successful, future research can evaluate the utility of machine learning and CAD in a variety of POCUS techniques in the ED, such as diagnostic POCUS of the abdomen and heart - potentially allowing faster and earlier treatment. These AI enhancements of traditional ultrasounds may greatly improve the speed and accuracy of diagnosis in busy EDs.
Co Author/Co-Investigator Names/Professional Title: Tushank Chadha, UC Irvine Emergency Medicine Research Associate Bhavesh H. Patel, UC Irvine Emergency Medicine Research Associate